Skip to content
AskVault Docs
Try Free →

AI safety commitments

Written by , Principal Architect, Networks and Tech, AskVault · Reviewed by Allison Chan
Last updated: · 5 min read

The five commitments

  1. Source-cited answers. Every bot answer cites a verifiable source. RAG architecture grounds responses; no free-form hallucination.
  2. No customer-data training. Your content and conversations are never used to train shared models.
  3. Audience-tag enforcement. Sensitive content scoped to verified visitors only.
  4. Kill switches per skill. Hard caps the LLM cannot override.
  5. Mandatory escalation paths. Every conversation has a path to a human.

Source-citation as anti-hallucination

Our primary defense against hallucination:

  • Retrieval before generation. Bot retrieves relevant chunks first.
  • Generation grounded in those chunks via system prompt.
  • Citation surface every answer with URL.
  • Strict mode prompts to refuse if no chunks match.

Our design goal is to keep hallucinations low. Most "wrong" answers reflect knowledge gaps in your content, not fabrication. Verify high-stakes answers via the source citations.

No customer-data training

A written commitment:

  • Your content never used to train shared models.
  • Your conversations never used.
  • LLM providers we route through have "not for training" flags set.

No opt-in alternative exists. If we ever introduced one, it would be explicit, off-by-default, revocable.

Audience-tag enforcement

To prevent leakage:

  • Every retrieval filters by audience.
  • The LLM never sees unauthorized content.
  • No prompt injection bypasses the filter (it operates pre-LLM).

Tested rigorously; we treat cross-audience leakage as P0 priority.

Skill kill switches

Some skills have hard limits beyond LLM control:

  • discount_negotiator. 15% discount hard cap, enforced per negotiation.
  • subscription_manager. ₹10,000/day refund cap per workspace.
  • collect_lead. Once per conversation.

Enforced at the policy layer, not in the prompt. LLM "decides" to offer 50% off; policy rejects.

Mandatory escalation

Every conversation surface has a path to a human:

  • "Talk to a human" trigger in widget and channel UIs.
  • escalate_to_human skill auto-triggers on frustration.
  • Bot retry-then-escalate on repeated misunderstandings.

No bot-only mode where customers feel trapped.

Bias and fairness

Best-effort:

  • Bot inherits LLM provider's bias in worst case.
  • Audience tags prevent serving sensitive content to wrong segments.
  • Content moderation filters extreme outputs.
  • Continuous monitoring of CSAT, complaints.

No bot is perfectly unbiased. We aim for fairness, monitor for issues.

Privacy

See data handling:

  • Encrypted at rest (AES-256).
  • TLS 1.3 in transit.
  • Workspace isolation at every layer.
  • GDPR data deletion within 30 days of request.

Red-team testing

Adversarial testing ahead of major releases:

  • Prompt injection attempts. Tested.
  • Audience-leak attempts. Continuous.
  • Skill-bound bypass attempts. Tested.
  • Findings remediated before next release.

We don't yet run this on a fixed public cadence or publish findings in a transparency report — ask us directly if you need specifics for a security review.

What we don't promise

Honesty:

  • Zero hallucination. Impossible. Aim for under 2%.
  • Perfect fairness. Best-effort; LLM bias residual.
  • Catching every prompt injection. Defense-in-depth; not invincible.
  • AGI safety. We deploy current models with current guardrails.

Be skeptical of vendors who promise more.

Limits

  • Hallucination rate. Target under 2% on factual questions.
  • Audience-tag enforcement. Architectural; verified per release.
  • Escalation latency. Not an SLA — the handoff trigger fires immediately, but time to a human response depends on agent availability and office hours.
  • Red-team cadence. Ad hoc, tied to major releases; not yet a fixed public schedule.
  • Audit retention. 365 days, flat across all plans.

Common pitfalls

Trusting bot uncritically. Verify high-stakes answers via citations.

Assuming all safety problems solved. Bot is one layer; humans review.

Removing human-escalation path. Required; don't disable.

FAQ

Will my bot get safer over time?

Yes. We continuously improve guardrails. Your bot inherits.

Can I see the system prompts?

Yes for your workspace under AI Config. Provider-side prompts not exposed (industry norm).

What if my bot gives a wrong answer?

Customers can thumbs-down. Agents can revise-and-train. Iterative improvement.

Was this page helpful?